Deep Learning Playground
Deconstruct neural networks to understand how they compute predictions. Step through layers, inspect weights and activations, see activation functions in action, and track the flow of forward and backward propagation — all in your browser.
Neural Network Forward Pass
Step through a fully-connected neural network layer by layer. Watch how input values get multiplied by weights, summed with biases, and transformed by activation functions to produce the final output. Configure the number of layers and neurons.
Backpropagation
Visualize how gradients flow backwards through a network during training. See the chain rule in action as the algorithm computes partial derivatives layer by layer, updating weights to minimize the loss function.
Activation Functions
Compare ReLU, Sigmoid, Tanh, Leaky ReLU, and more side by side. See their curves, derivatives, and how they affect neuron outputs. Understand why non-linearity is essential for deep networks to learn complex patterns.
CNN Operations
Watch convolution, pooling, and padding operations unfold step by step on input matrices. Adjust kernel size, stride, and padding to see how each parameter changes the output feature map dimensions and values.
Dropout Layer
Visualize how dropout randomly deactivates neurons during training to prevent overfitting. See which neurons get masked in each forward pass, and understand the scaling factor applied during inference.
Normalization
Explore batch normalization, layer normalization, and other techniques that stabilize training. See how raw activations get normalized, scaled, and shifted, and why this dramatically speeds up convergence.
CNN Feature Map Explorer
Upload an image and watch how different convolutional filters extract features like edges, textures, and patterns. Explore what each layer of a CNN 'sees' and how representations become increasingly abstract deeper in the network.
CNN Architecture Visualizer
Build and visualize the full architecture of a convolutional neural network. Add convolutional, pooling, and fully-connected layers, and see how tensor dimensions transform through the entire pipeline from input to output.
What You'll Learn
How Neural Networks Compute
Trace the exact math behind forward propagation — from weighted sums and biases to activation outputs across multiple layers.
Training with Backpropagation
Understand how the chain rule enables gradient computation and how weight updates reduce loss over training iterations.
Convolutional Neural Networks
See how convolution, pooling, and feature maps work together to build hierarchical visual representations from raw pixels.
Regularization Techniques
Explore dropout, normalization, and other methods that prevent overfitting and stabilize training in deep networks.
Ready to Explore Neural Networks?
Start with the Forward Pass visualizer — step through a neural network one layer at a time and see exactly how inputs become outputs.
Launch Forward Pass